# Finding 085: Copilot/Bing repeated the plain head-signal failed fetch without target evidence

## Date

2026-07-02

## Status

Published

## Summary

Copilot/Bing was prompted in a fresh Temporary chat with
`manual-client-copilot-bing-plain-head-signal-repeat-20260702-001-p38`, a repeat
of the plain site-owner review prompt against the
`/lab/reading/head-signal-isolation` fixture. The prompt asked Copilot/Bing to
open the target URL and briefly review the page, without asking for source
areas, page-head metadata, structured data, code-like tokens, marker values,
or hidden fields.

Copilot/Bing returned `fetched:false`, `pages_opened:0`, and evidence quote
`Failed to get web content.` A bounded origin review found no exact
target-page hit and no related `/`, `/robots.txt`, or sitemap activity inside
the prompt window. This repeats the p37 Copilot/Bing failed-fetch answer and
clean target no-hit pattern under the same plain site-owner review wording.

## What does this mean?

For site owners, SEO/AEO teams, publishers, and researchers, this repeat shows that a mainstream AI client can answer that it could not fetch a page even when the same lab fixture is reachable and other clients have read it. The important part is not the model's wording alone: the server logs also showed no matching page read during the prompt window, so this run should be treated as a real client-specific no-hit rather than a hidden successful visit.

## Method

- Controlled-browser task:
  `research/manual-client-runs/browser-tasks/manual-client-copilot-bing-plain-head-signal-repeat-20260702-001-p38.browser-task.json`.
- Prompt packet:
  `research/manual-client-runs/manual-client-copilot-bing-plain-head-signal-repeat-20260702-001.prompts.json`.
- Response artifact:
  `research/manual-client-runs/browser-tasks/responses/manual-client-copilot-bing-plain-head-signal-repeat-20260702-001-p38.response.json`.
- Logged answer packet:
  `research/manual-client-runs/manual-client-copilot-bing-plain-head-signal-repeat-20260702-001.answers.json`.
- Prompt window:
  `2026-07-02T15:20:19.869Z` to `2026-07-02T15:20:44.978Z`.

The prompt was submitted in a fresh Copilot Temporary chat in the logged-in
OpenClaw Chrome profile. The visible UI showed the `kai` account, Free Plan,
Temporary mode, and `Smart` response mode. An initial unsent tab accidentally
opened Copilot voice mode and showed a microphone permission screen; no test
prompt was submitted there. The exact p38 prompt was then submitted in a new
fresh Temporary chat.

## Evidence

| Evidence | Result |
|---|---|
| Model answer | `fetched:false`, `pages_opened:0`; `Failed to get web content.` |
| Copilot/Bing UI state | Fresh Temporary chat; visible account `kai`; Free Plan; response mode `Smart`. |
| Direct-origin target-page event | None for `/lab/reading/head-signal-isolation?id=manual-client-copilot-bing-plain-head-signal-repeat-20260702-001-p38` inside the bounded prompt window. |
| Ancillary origin activity | None for related `/`, `/robots.txt`, sitemap, or exact attempt-id paths inside the bounded prompt window. |

```json
{
  "timestampWindow": {
    "startedAt": "2026-07-02T15:20:19.869Z",
    "endedAt": "2026-07-02T15:20:44.978Z"
  },
  "confirmedHitFromPrompt": false,
  "rawEventIds": [],
  "ancillaryOriginActivity": {
    "present": false,
    "paths": [],
    "rawEventIds": []
  }
}
```

A wider +/-120s raw-event review found two root-path requests from
`::ffff:161.35.3.45` about nine seconds after the copied answer ended. They
had no attempt id, no target path, and generic browser user agents, so they
were reviewed as not prompt-confirming target evidence.

## Interpretation

This p38 repeat is a clean target no-hit. It supports the narrow claim that
Copilot/Bing again reported target-page retrieval failure under the same plain
head-signal review wording, and the lab did not record any exact target-page
hit or related ancillary origin activity inside the bounded prompt window.

Compared with Finding 082, the answer-side result and origin-side target
state are stable: both p37 and p38 returned `Failed to get web content.` and
both bounded reviews found no exact target-page hit and no related root,
robots, or sitemap activity inside the prompt window. The result should still
be scoped to one account, plan, fixture, prompt family, and time window.

## Limitations

- This is one Copilot/Bing repeat from one account, plan, client mode, and
  time window.
- The prompt supplied the exact target URL, so this measures direct URL
  opening rather than independent discovery.
- The answer and origin log agree on no target-page retrieval, but they do
  not explain whether Copilot/Bing skipped retrieval, used cached routing
  state, failed before contacting the lab, or contacted an unobserved
  intermediate service.
- The lack of related origin activity is bounded to the prompt window and the
  retained local event log.
- An initial unsent tab triggered Copilot voice mode and a microphone
  permission screen; the actual p38 prompt was submitted only after opening a
  new fresh Temporary chat.

## Publication Thesis Verification

- Thesis: In the p38 plain site-owner review repeat, Copilot/Bing returned a
  failed-fetch answer and the lab recorded no exact target-page hit or related
  ancillary origin activity inside the bounded prompt window.
- Source: The p38 browser-task artifact, response artifact, logged answer
  packet, Copilot UI snapshot context, and direct-origin raw event review for
  `2026-07-02T15:20:19.869Z` to `2026-07-02T15:20:44.978Z`.
- Method: Compare the copied Copilot/Bing answer and visible browser state
  with the exact attempt id, prompt code, timestamp window, and local
  server-side events for the target path and nearby ancillary paths.
- Bias: Single-account, single-run repeat evidence from a synthetic fixture.
  The prompt is plain relative to source-area prompts but still names
  AI-assistant readability and an exact lab URL.
- Consensus: Consistent with Finding 082 on failed target-page retrieval and
  clean target no-hit state. It also contrasts with the confirmed-hit
  ChatGPT and Gemini p37 runs against the same fixture.
- Invalidation: A corrected origin review finding an exact target-page event
  or related Copilot/Bing ancillary event inside the prompt window, or a
  response-artifact mismatch showing that the answer belonged to a different
  attempt, would weaken this finding.
- Verdict: Supported for the p38 run. It documents a registered clean target
  no-hit with a copied Copilot/Bing failed-fetch answer.
- Additional tests suggested: publish a p37/p38 no-hit-client comparison that
  separates Perplexity answer stability, Perplexity ancillary-origin
  variability, and Copilot/Bing's repeated clean no-hit state.

## Next steps

- Compare Perplexity and Copilot/Bing p37/p38 repeats before claiming whether
  no-hit outcomes are stable for this prompt family.
- Consider a later mixed-order repeat packet if the lab needs stronger
  evidence about account, timing, or URL-order effects.